Mistral Small 4 vs Qwen3.8 27B
Qwen3.8 27B passed more tasks on the AIGROW small-business benchmark and Mistral Small 4 has the lower list price, so pick Qwen3.8 27B for accuracy and Mistral Small 4 for cost. Mistral Small 4 passed 4 of 20 tasks and costs $0.15 in and $0.60 out per million tokens; Qwen3.8 27B passed 16 and costs $0.50 in and $3 out.
AIGROW credit covers Claude Opus 5.5, Claude Fable 5.2, GPT-6.2 Astra, Claude Sonnet 5.5, Claude Haiku 5.5, GPT-6.2 Sol, GPT-6.1 Terra, GPT-6 Luna, Grok 4.7, Gemini 4 Argon, Kimi K3.1 Pro, Gemini 4 Hydrogen, Gemini 3.8 Flash, Claude Fable 5.1 (previous version), GPT-6 Astra (previous version), GPT-6.1 Sol (previous version) and Qwen3.8 Max (previous version). See API credit
$25 a month brings $110 of API credit and cancels any time; $10 once buys $40. For long documents, Qwen3.8 27B reads up to 1M tokens at once and Mistral Small 4 up to 256K. US dollars, standard rate, shortest context band. Cost per pass from the AIGROW small-business benchmark. Prices checked 18 November 2026.
- Mistral Small 4
- $1.33
- Tasks passed
- 4 of 20
- Qwen3.8 27B
- $7.30
- Tasks passed
- 16 of 20
Open-weight budget APIs, with Mistral Small 4 at 30% of Qwen3.8 27B's hosted input price and a fifth of its output. Mistral Small 4 costs $1.33 per 1,000 passing answers on the AIGROW small-business benchmark, Qwen3.8 27B $7.30. 1,000 support replies cost $2.70 on Qwen3.8 27B and $0.69 on Mistral Small 4, so Qwen3.8 27B costs 3.9 times as much.
What each one costs
The API price per million tokens and the cheapest app plan with each model, from the makers’ own pages. Each name opens that model’s full price page.
- Mistral Small 4
- On the Mistral API, Mistral Small 4 costs $0.15 per million input tokens and $0.60 per million output tokens. Vibe has a free plan and paid plans from $14.99/mo (Pro); the plan list does not name Mistral Small 4. Plan prices read 22 October 2026.
- Qwen3.8 27B
- On the Alibaba (Qwen) API, Qwen3.8 27B costs $0.50 per million input tokens and $3 per million output tokens.
Side by side
Each maker’s published rate for its standard tier. Cached input is what a repeated prompt prefix costs once the provider has stored it.
| Fact | Mistral Small 4 | Qwen3.8 27B |
|---|---|---|
| Maker | Mistral | Alibaba (Qwen) |
| Tier | Small and fast | Small and fast |
| Input, per M tokens | $0.15 | $0.50 |
| Output, per M tokens | $0.60 | $3.00 |
| Cached input | $0.015 | Not published |
| Context window | 256K | 1M |
| Released | 16 Mar 2026 | 17 Aug 2026 |
| Weights | Open weights | Open weights |
What a job costs on each
1,000 runs of each job at the two list prices. The gap moves from job to job because the models price input, output and cached context differently.
| Job | Mistral Small 4 | Qwen3.8 27B | Cheaper |
|---|---|---|---|
| Support reply3,000 in, 400 out. | $0.69 | $2.70 | Mistral Small 4, 74% less |
| Document summary25,000 in, 1,000 out. | $4.35 | $15.50 | Mistral Small 4, 72% less |
| Agent task80,000 in (70,000 cached), 3,000 out. | $4.35 | $49.00 | Mistral Small 4, 91% less |
The token counts are ours, chosen to look like real work; the prices are the makers’. A model with no published cache price bills the agent task’s repeated context at its full input rate.
The small print
What each base rate leaves out, from the makers’ own pages: surcharges, discounts, promotions and cache pricing.
- Mistral Small 4
- Mistral Small 4. 119B total, 6.5B active parameters, Apache 2.0. Batch 50% off. Mistral’s pricing page
- Qwen3.8 27B
- Open-source model with a first-party hosted price (listed under 'Qwen (open source)'). International deployment, one tier up to 1M input tokens. Context taken from the top pricing tier (1M). Alibaba (Qwen)’s pricing page
Other comparisons
Every head to head on the list that involves Mistral Small 4 or Qwen3.8 27B.
- Claude Haiku 5.5 vs Qwen3.8 27B
- GLM-5.3-Flash vs Qwen3.8 27B
- Qwen3.8 27B vs Gemini 3.8 Flash
- Qwen3.8 Flash vs Qwen3.8 27B
Comparisons with an earlier generation
Compare Mistral Small 4 and Qwen3.8 27B with 62 other models
The price list sets every model’s input and output price side by side, with what a support reply, a document summary and an agent task cost on each.
Prices checked 18 November 2026